def LassoRegionOptimizer(str_f_iosg, str_f_st, str_refgem,
str_of_seq, str_of_fesrep, str_of_md,
int_regn_start=0, int_regn_end=100, int_regn_step=5,
cv=5, niter=1000, njob=1):
int_ups_regn = np.arange(int_regn_start, int_regn_end+1, int_regn_step)
int_dws_regn = np.where((int_ups_regn-3)>0, int_ups_regn-3, 0)
arr_r2 = np.zeros_like(int_ups_regn, dtype=np.double)
dfm_st = pd.read_csv(str_f_st, sep='\t', index_col=None)
dfm_st = dfm_st[[0,-1]]
for (i,(int_ups, int_dws)) in enumerate(zip(int_ups_regn, int_dws_regn)):
dfm_x = ExtractSeqFeature(str_f_iosg, str_refgem, int_ups, int_dws)
dfm_y = pd.merge(dfm_x, dfm_st, on='sgID')
dfm_x = dfm_y.ix[:, 1:-1]
dfm_y = dfm_y[[-1]]
x = np.array(dfm_x, dtype=np.double)
y = np.array(dfm_y, dtype=np.double).ravel()
arr_r2[i] = ml.LassoSelector(x, y, cv, niter, njob).r2
int_ups = int_ups_regn[arr_r2.argmax()]
int_dws = int_dws_regn[arr_r2.argmax()]
dfm_x = ExtractSeqFeature(str_f_iosg, str_refgem, int_ups, int_dws)
dfm_x.to_csv(str_of_seq, sep='\t', index=None)
dfm_y = pd.merge(dfm_x, dfm_st, on='sgID')
dfm_x = dfm_y.ix[:, 1:-1]
dfm_y = dfm_y[[-1]]
x = np.array(dfm_x, dtype=np.double)
y = np.array(dfm_y, dtype=np.double).ravel()
md = ml.LassoSelector(x, y, cv, niter, njob)
lst_fs = [i for i in dfm_x.columns[md.idx]]
fr = Fr(fs=lst_fs, idx=md.idx, ups=int_ups, dws=int_dws, niter=niter,
cv=cv, cor=md.cor, r2=md.r2)
__SaveFeatureReport(fr, str_of_fesrep)
mdl = dict(med='lasso', model=md.model, idx=md.idx,
ups=int_ups, dws=int_dws)
ml.SaveModel(mdl, str_of_md)
def LogitRegionOptimizer(str_f_iosg, str_f_st, str_refgem,
str_of_seq, str_of_fesrep, str_of_md,
int_regn_start=0, int_regn_end=100, int_regn_step=5,
cv=5, njob=1):
int_ups_regn = np.arange(int_regn_start, int_regn_end+1, int_regn_step)
int_dws_regn = np.where((int_ups_regn-3)>0, int_ups_regn-3, 0)
arr_accu = np.zeros_like(int_ups_regn, dtype=np.double)
dfm_st = pd.read_csv(str_f_st, sep='\t', index_col=None)
dfm_st = dfm_st[[0,-1]]
for (i,(int_ups, int_dws)) in enumerate(zip(int_ups_regn, int_dws_regn)):
dfm_x = ExtractSeqFeature(str_f_iosg, str_refgem, int_ups, int_dws)
dfm_y = pd.merge(dfm_x, dfm_st, on='sgID')
dfm_x = dfm_y.ix[:, 1:-1]
dfm_y = dfm_y[[-1]]
x = np.array(dfm_x, dtype=np.double)
y = np.array(dfm_y, dtype=np.int)
arr_accu[i] = ml.LogitSelector(x, y, cv, njob).accu
int_ups = int_ups_regn[arr_accu.argmax()]
int_dws = int_dws_regn[arr_accu.argmax()]
dfm_x = ExtractSeqFeature(str_f_iosg, str_refgem, int_ups, int_dws)
dfm_x.to_csv(str_of_seq, sep='\t', index=None)
dfm_y = pd.merge(dfm_x, dfm_st, on='sgID')
dfm_x = dfm_y.ix[:, 1:-1]
dfm_y = dfm_y[[-1]]
x = np.array(dfm_x, dtype=np.double)
y = np.array(dfm_y, dtype=np.int)
mdc = ml.LogitSelector(x, y, cv, njob)
lst_fs = [i for i in dfm_x.columns[mdc.idx]]
frc = Frc(fs=lst_fs, ups=int_ups, dws=int_dws,
cv=cv, accu=mdc.accu, prec=mdc.prec, rec=mdc.rec,
f1=mdc.f1, au=mdc.au)
__SaveFeatureReport(frc, str_of_fesrep)
mdlc = dict(med='logit', model=mdc.model, idx=mdc.idx,
ups=int_ups, dws=int_dws)
ml.SaveModel(mdlc, str_of_md)
def MtLassoFeatureSelection(lst_sg, lst_st, lst_ref, lst_seq,
str_of_fesrep, int_ups=30, int_dws=27, flt_lmd=0.5):
lst_x = []
lst_y = []
ind = [0]
lst_header = []
for i in xrange(len(lst_sg)):
dfm = ExtractSeqFeature(lst_sg[i], lst_ref[i], int_ups, int_dws)
dfm.to_csv(lst_seq[i], sep='\t', index=None)
dfm_y = pd.read_csv(lst_st[i], sep='\t', index_col=None)[[0,-1]]
dfm = pd.merge(dfm, dfm_y, on='sgID')
x = np.array(dfm.ix[:,1:-1], dtype=np.double)
y = np.array(dfm[[-1]], dtype=np.double)
lst_x.append(x)
lst_y.append(y)
ind.append(dfm.shape[0])
if i == 0:
lst_header = dfm.columns[1:-1].tolist()
x = np.double(np.vstack(lst_x))
y = np.double(np.vstack(lst_y))
ind = np.cumsum(ind)
dict_opts = dict(init=2, rFlag=1, rsL2=0, ind=ind, nFlag=2)
idx = ml.MtLassoSelector(x, y, flt_lmd, dict_opts)
lst_fs = [lst_header[i] for i in idx]
fr = Fr(fs=lst_fs, ups=int_ups, dws=int_dws)
__SaveFeatureReport(fr, str_of_fesrep)
def FeatureEval(str_f_iosg, str_f_md, str_refgem, str_of_seq, str_of_st):
mdl = ml.LoadModel(str_f_md)
dfm = ExtractSeqFeature(str_f_iosg, str_refgem, mdl['ups'], mdl['dws'])
dfm.to_csv(str_of_seq, sep='\t', index=None)
x = np.array(dfm.ix[:,1:], dtype=np.double)
y = None
if mdl['med'] == 'lasso':
y = ml.LassoEval(mdl['model'], x, mdl['idx'])
elif mdl['med'] == 'logit':
y = ml.LogitEval(mdl['model'], x, mdl['idx'])
dfm['score'] = y
dfm_y = dfm[[0,-1]]
dfm_y.to_csv(str_of_st, sep='\t', index=None)
def LogitFeatureSelection(
str_f_iosg,
str_f_st,
str_refgem,
str_of_seq,
str_of_fesrep,
str_of_md,
int_ups=30,
int_dws=27,
cv=5,
niter=1000,
njob=1,
):
dfm_x = ExtractSeqFeature(str_f_iosg, str_refgem, int_ups, int_dws)
dfm_x.to_csv(str_of_seq, sep="\t", index=None)
dfm_y = pd.read_csv(str_f_st, sep="\t", index_col=None)
dfm_y = dfm_y[[0, -1]]
dfm_y = pd.merge(dfm_x, dfm_y, on="sgID")
dfm_x = dfm_y.ix[:, 1:-1]
dfm_y = dfm_y[[-1]]
x = np.array(dfm_x, dtype=np.double)
y = np.array(dfm_y, dtype=np.int)
mdc = ml.LogitSelector(x, y, cv, niter, njob)
lst_fs = [i for i in dfm_x.columns[mdc.idx]]
frc = Frc(
fs=lst_fs,
idx=mdc.idx,
ups=int_ups,
dws=int_dws,
niter=niter,
cv=cv,
accu=mdc.accu,
prec=mdc.prec,
rec=mdc.rec,
f1=mdc.f1,
au=mdc.au,
)
__SaveFeatureReport(frc, str_of_fesrep)
mdlc = dict(med="logit", model=mdc.model, idx=mdc.idx, ups=int_ups, dws=int_dws)
ml.SaveModel(mdlc, str_of_md)
def LassoFeatureSelection(
str_f_iosg, str_f_st, str_refgem, str_of_seq, str_of_fesrep, str_of_md, int_ups=30, int_dws=27, cv=5, njob=1
):
dfm_x = ExtractSeqFeature(str_f_iosg, str_refgem, int_ups, int_dws)
dfm_x.to_csv(str_of_seq, sep="\t", index=None)
dfm_y = pd.read_csv(str_f_st, sep="\t", index_col=None)
dfm_y = dfm_y[[0, -1]]
dfm_y = pd.merge(dfm_x, dfm_y, on="sgID")
dfm_x = dfm_y.ix[:, 1:-1]
dfm_y = dfm_y[[-1]]
x = np.array(dfm_x, dtype=np.double)
y = np.array(dfm_y, dtype=np.double).ravel()
md = ml.LassoSelector(x, y, cv, njob)
lst_fs = [i for i in dfm_x.columns[md.idx]]
fr = Fr(fs=lst_fs, ups=int_ups, dws=int_dws, cv=cv, cor=md.cor, r2=md.r2)
__SaveFeatureReport(fr, str_of_fesrep)
mdl = dict(med="lasso", model=md.model, idx=md.idx, ups=int_ups, dws=int_dws)
ml.SaveModel(mdl, str_of_md)
def FeatureEval(str_f_iosg, str_f_md, str_refgem, str_of_seq, str_of_st):
mdl = ml.LoadModel(str_f_md)
dfm = ExtractSeqFeature(str_f_iosg, str_refgem, mdl["ups"], mdl["dws"])
dfm.to_csv(str_of_seq, sep="\t", index=None)
x = np.array(dfm.ix[:, 1:], dtype=np.double)
y = None
if mdl["med"] == "lasso":
y = ml.LassoEval(mdl["model"], x, mdl["idx"])
elif mdl["med"] == "logit":
y = ml.LogitEval(mdl["model"], x, mdl["idx"])
dfm["score"] = y
dfm_y = dfm[[0, -1]]
dfm_iosg = pd.read_csv(str_f_iosg, sep="\t", header=None, index_col=None)
dfm_iosg.columns = ["sgID", "chrom", "strand", "sbeg", "send", "qseq", "c_site"]
dfm = pd.merge(dfm_iosg, dfm_y, on="sgID")
dfm.drop("c_site", axis=1, inplace=True)
dfm.sort_index(by=["score"], ascending=False, inplace=True)
dfm.to_csv(str_of_st, sep="\t", index=None)